Documentation Hub ​

Welcome to the ARG v1.1 documentation. This page helps you choose the right entry point depending on whether you are designing the protocol, the taxonomy layer, governance rules, or operational build practices.

Start here ​

• New to ARG? Begin with Introduction for scope, goals, and the core architectural rationale.
• Want the protocol itself? Go to ARG Core.

The architecture, by responsibility ​

1) Protocol & execution logic ​

ARG Core — This is the reference for:

• the online inference loop (routing → traversal → action/info/memory),
• the offline refinement loop (split/merge/edges/labels),
• controlled evolution and lifecycle rules.

If you are implementing the full agent pipeline, this is your foundation.

2) Taxonomy-safe label arbitration under latency constraints ​

Context Weaver — Use this section if you need:

• a taxonomy-first label selection system,
• a Distilled-first cascade,
• deterministic validation,
• safe abstention and cold-start handling.

The Context Weaver is the semantic stabilizer that protects both routing accuracy and long-term memory quality.

3) Governance & safety ​

Policy Manager — This is your kernel of control, independent from the graph. It defines:

• Pre-Check gating before routing,
• constraint injection for retrieval, traversal and scoring,
• Post-Check before response and memory writes,
• strict controls for RBAC/ABAC and memory safety.

4) Ontology construction, audits, and modeling rules ​

Guides — This page consolidates the practical manuals:

• how to build clusters, labels, nodes, and edges correctly,
• how to avoid structural drift and keep memory stable as the graph evolves,
• how to run pre-IA audits that catch foundational errors early,
• how to design governed action agents (when to act, how to model actions, how policy gates execution),
• how to build bounded retrieval agents (routing, landing points, traversal budgets, context bundles),
• how to implement short- and long-term memory (episodic writes, canonicalization, deduplication, offline consolidation).

If you're onboarding a domain team or bootstrapping a new vertical, start here after the Intro.

Recommended reading paths ​

Implementation path (engineering) ​

1. Introduction
2. ARG Core
3. Context Weaver
4. Policy Manager
5. Guides

Modeling path (domain + knowledge design) ​

1. Introduction
2. Guides
3. Context Weaver
4. ARG Core

Governance path (compliance + security) ​

1. Introduction
2. Policy Manager
3. ARG Core

What ARG standardizes ​

Across all pages, ARG v1.1 formalizes:

• a taxonomy-driven operational ontology,
• a leaf-oriented reasoning graph (nodes as executable/knowledge leaves),
• bounded vector acceleration (fast but non-authoritative),
• a dual memory model with safe offline consolidation,
• measurable outcomes that never confuse silence with success.

Where to go next ​

• Scope and rationale → Introduction
• End-to-end protocol → ARG Core
• Label arbitration + latency design → Context Weaver
• Governance kernel → Policy Manager
• Construction & audit best practices → Guides